Claude Haiku 5.5 Pricing: $0.10 / $0.50, 90% Below Haiku 4.5
Anthropic's new small model matches GPT-6 Luna on price, beats it on Anthropic's benchmarks, and comes with a catch for prompts over 100,000 tokens.
Anthropic released Claude Haiku 5.5 on 7 October 2026 at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. That is 90% below Claude Haiku 4.5's $1 / $5. Cached input costs $0.01 per million, and batch requests are half price. Anthropic says that, on average, it costs about 75% less to run than Haiku 4.5.
On the same day, Anthropic halved the price of cached input on Claude Sonnet 5.5, from $0.20 to $0.10 per million tokens.
Claude Haiku 5.5 at a glance
| API model ID | claude-haiku-5-5 |
|---|---|
| Input | $0.10 per 1M tokens ($0.50 for prompts over 100K) |
| Output | $0.50 per 1M tokens ($2.50 for prompts over 100K) |
| Cached input | $0.01 per 1M tokens ($0.05 for prompts over 100K) |
| Cache writes | $0.125 per 1M tokens ($0.625 for prompts over 100K) |
| Batch | 50% off: $0.05 / $0.25 per 1M tokens |
| Context window | 1M tokens |
| Maximum output | 128K tokens |
It is the first Haiku model with an adjustable effort setting, so you can trade cost against quality as with the larger Claude models. The live figures are on our Claude Haiku pricing page.
The 100,000-token catch
Haiku 5.5 is priced by prompt length. If a request's prompt is over 100,000 tokens, the whole request is billed at five times the rate: $0.50 / $2.50, with cache reads at $0.05. Cached tokens count towards the 100,000. Anthropic's larger models charge the same rate across their full 1M context, and GPT-6 Luna only raises its price above 272,000 tokens, and then by less (2× input, 1.5× output).
For most uses this doesn't matter: Anthropic says about 90% of requests to its previous Haiku model had prompts under 100,000 tokens. It does matter for long agent runs and large documents. Our agent cost calculator applies the higher rate to every step whose context passes the threshold.
How it compares with the rest of the Claude line-up
| Model | Input / output per 1M | Cached input |
|---|---|---|
| Claude Fable 5.1 | $10.00 / $50.00 | $0.25 |
| Claude Opus 5.5 | $4.00 / $20.00 | $0.20 |
| Claude Sonnet 5.5 | $2.00 / $10.00 | $0.10 |
| Claude Haiku 5.5 | $0.10 / $0.50 | $0.01 |
| Claude Haiku 4.5 | $1.00 / $5.00 | $0.10 |
Haiku 5.5 costs a twentieth of Sonnet 5.5 on input and output. Anthropic positions it for high-volume work such as classification, summaries, extraction and customer support, and as a subagent under Opus or Sonnet in coding tools.
Performance
These are Anthropic's published results, not our own tests.
| Benchmark | Haiku 5.5 | Haiku 4.5 | GPT-6 Luna | Sonnet 5.5 |
|---|---|---|---|---|
| GDPval-AA v2.1 (knowledge work, Elo) | 1620 | 735 | 1437 | 1840 |
| OSWorld 2.1, offline subset (computer use) | 72.4% | 15.7% | 48.9% | 83.9% |
| Terminal-Bench 4.0 (agentic coding) | 39.2% | 0.0% | 16.4% | 70.6% |
| Chartography (charts, no tools) | 46.4% | 6.4% | 29.1% | 61.6% |
| Humanity's Last Exam (no tools) | 45.9% | 10.2% | — | 56.9% |
The jump from Haiku 4.5 is large on every measure, and on these benchmarks Haiku 5.5 scores well above GPT-6 Luna at the same price. Anthropic is clear that Sonnet 5.5 and Opus 5.5 remain the better choice for complex agentic coding.
On the independent Artificial Analysis Intelligence Index, Haiku 5.5 scored 43 on 13 October, second of the 180 models in its price class, and was measured at 235 output tokens per second, faster than any Claude model before it.
What it costs in practice
For a high-volume classification or extraction job, say 100M input and 10M output tokens a month with short prompts, Haiku 5.5 costs $15.00, against $150 on Haiku 4.5, $15.00 on GPT-6 Luna and $113 on Gemini 3.8 Flash.
With half of the input served from cache, which is common for a fixed system prompt, Haiku 5.5 comes to $10.50. Compare Haiku 5.5 with GPT-6 Luna or with Gemini 3.8 Flash.
Also announced
- Sonnet 5.5 cache price halved: cached input now costs $0.10 per million tokens, down from $0.20. Anthropic says this makes Sonnet 5.5 about 20% cheaper on most agent work.
- API credits for subscribers: Claude Max subscribers get monthly credits for the Claude Platform API: $100 on Max 5x and $200 on Max 20x. Team plans get up to $500, pooled across users. That changes the subscription versus API sums for anyone building their own tools.
Should you switch?
- From Haiku 4.5: yes, for almost everything. It is cheaper per token and far stronger on Anthropic's benchmarks. Check any workloads with prompts over 100,000 tokens, which cost five times more. Haiku 4.5 stays available at $1 / $5 while you test.
- From GPT-6 Luna: the per-token price is identical, so test both on your own tasks. Luna is cheaper for prompts between 100,000 and 272,000 tokens.
- From Sonnet 5.5, for simple or high-volume steps: routing summaries, classification and subagent work to Haiku 5.5 can cut those costs by around 95%.
- If you work in security: Haiku 5.5's cybersecurity safeguards are stricter than Haiku 4.5's. Defensive tasks are allowed, but penetration testing is blocked.
Sources: Anthropic pricing, Anthropic announcement, Artificial Analysis.